CORRELATION ENGINE FOR GENERATING ANONYMOUS CORRELATIONS BETWEEN PUBLICATION-RESTRICTED DATA AND PERSONAL ATTRIBUTE DATA

• US 20100036884A1
• Filed: 08/06/2009
• Published: 02/11/2010
• Est. Priority Date: 08/08/2008
• Status: Abandoned Application

• Current Assignee: Arcametrics Systems, Inc. (Adroit Digital)
• Original Assignee: Arcametrics Systems, Inc. (Adroit Digital)
• Inventors: Brown, Robert G.
• Application Number:
  US12/536,765
• Publication Number:
  US 20100036884A1
• Time in Patent Office: Days
• US Class Current: N/A
• CPC Class Codes: G06Q 10/067 Enterprise or organisation modelling
• CPC Class Codes: G06Q 30/00 Commerce
• CPC Class Codes: G06Q 50/265 Personal security, identity or safety

Abstract

A correlation engine apparatus includes a network interface and a processor, wherein the correlation engine is configured to receive publication-restricted data and non-publication-restricted data and generate correlations useable for predictive models, wherein no trace of any personal identifying information (PII) in the publication-restricted data exists in the correlations.

Claims

1. A correlation engine apparatus comprising

a network interface;

and

a processor;

wherein the correlation engine is configured to receive publication-restricted data and non-publication-restricted data and generate correlations useable for predictive models, wherein no trace of any personal identifying information (PII) in the publication-restricted data exists in the correlations.

2. The correlation engine according to claim 1, wherein the non-publication-restricted data comprises attribute data of persons, the attribute data comprising demographic attributes of persons.

3. The correlation engine according to claim 1, further comprising an encoding/decoding device, the encoding/decoding device configured to decode the received publication-restricted data and non-publication-restricted data and to encode the generated correlations.

4. The correlation engine according to claim 1, further comprising the correlation engine being configured to generate an anonymous aggregated correlation map from the publication-restricted data.

5. The correlation engine according to claim 4, wherein the anonymous aggregated correlation map is useable as a Bayesian prior for computations estimating probabilities using transactional data.

6. The correlation engine according to claim 4, further comprising the correlation engine being configured to generate a reverse projection correlation map, the reverse projection correlation map being generated by combining the anonymous aggregated correlation map with the non-publication-restricted data based on selected closely matching transactional categories.

7. The correlation engine according to claim 6, wherein the reverse projection correlation map is useable as a Bayesian prior for strategic business purposes.

8. The correlation engine according to claim 6, wherein the reverse projection correlation map is useable to provide a best guess as to originating non-publication-restricted data without the use of PII based on specific transaction data.

9. The correlation engine according to claim 6, further comprising the correlation engine being configured to generate a unified correlation map from the anonymous aggregated correlation map and the reverse projection correlation map, the unified correlation map being generated quantitatively and objectively and non-heuristically.

10. The correlation engine according to claim 9, wherein the unified correlation map comprises a set of all Bayesian reductions of the anonymous aggregated correlation map and the reverse projection correlation map.

11. The correlation engine according to claim 1, further comprising the correlation engine being configured to generate a correlated predictive model.

12. The correlation engine according to claim 1, further comprising the correlation engine being configured to generate a multistage neural posterior predictive model.

13. The correlation engine according to claim 1, further comprising the correlation engine being configured to generate a Parzen-Bayes network predictive model.

14. A method for generating a predictive model comprising:

receiving transaction data;

generating personal identity information (PII) free transaction data by removing any personal identity information contained in the transaction data;

generating probability distributions for each transactional category of interest contained in the PII free transaction data;

generating joint and conditional probability distributions based on at least two transactional categories; and

generating a forward map predictive model P of the joint and conditional probability distributions.

15. The method according to 14, further comprising analyzing the generated joint and conditional probability distributions to determine whether the generated joint and conditional probability distributions are satisfactory and re-generating joint and conditional probability distributions based on the at least two transactional categories and at least one new transactional category when the generated joint and conditional probability distributions are not satisfactory.

16. A method for generating a predictive model comprising:

identifying matching transaction categories between personal identity information (PII) free transaction data and PII free demographic transaction data;

generating probability distributions for all transaction and demographic variables in the matching transaction categories;

generating joint and conditional probability distributions based on at least two transaction and demographic variables; and

generating a reverse map predictive model Q of conditional probabilities from the matching transaction categories back to the demographic variables.

17. The method according to claim 16, further comprising receiving transaction data and generating the PII free transaction data by removing any personal identity information contained in the transaction data.

18. The method according to claim 16, further comprising receiving demographic transaction data associated with at least one person and generating the PII free demographic transaction data by removing any personal identity information contained in the demographic transaction data.

19. The method according to 16, further comprising analyzing the generated joint and conditional probability distributions to determine whether the generated joint and conditional probability distributions are satisfactory and re-generating joint and conditional probability distributions based on the at least two transactional categories and demographic variables and at least one new transactional category when the generated joint and conditional probability distributions are not satisfactory.

20. A method for utilizing data in a publication-restricted database in a manner that avoids publication of personal identity information (PII) data, the method comprising:

(a) generating, from the data in the publication-restricted database a set of aggregated multidimensional matrices that represent one of a population frequency or estimated joint probability of individuals in that database participating in selected constellations of transactions or other behaviors;

(b) constructing predictive models that target propensity to participate in particular transactions or other behaviors represented in one or more of these linked joint probability constellations;

(c) deriving, from the set of joint probability constellations that represent strongly correlated transactions or other behaviors within the publication-restricted database, predictive models for additional, strongly correlated transactions or other behaviors distinct from the particular model constructed in (b); and

(d) utilizing this set of derived models as input in the construction of additional predictive models that target transactions or other behaviors linked by them,

wherein the construction of predictive models are enabled by correlating data in the publication-restricted database with non-publication-restricted data found in a separate and distinct database.

21. The method according to claim 20, further comprising:

analyzing the predictive models constructed to identify specific ranges of subsets of the input variables that are strongly associated with particular transactional or other behavioral constellations; and

transforming these identified ranges into “

business intelligence”

that can be used to further direct model generation and other business activity such as the creation of new products in the linked transactional or other behavioral categories,

wherein the construction of inverted projective maps are enabled from which common characteristics of individuals in particular transactional or behavioral groups can be deduced without the direct use of the publication-restricted information.